Seismic data analysis with dual-tree M-band wavelet transforms

Laurent Duval, Caroline Chaux, Stéphan Ker2007 · DOI

TL;DR

This paper introduces a novel seismic data processing technique using dual-tree M-band wavelet transforms. This method effectively tackles challenges in seismic data analysis, particularly for denoising and local slope estimation, by combining enhanced directional analysis with a manageable level of redundancy.

What Problem Does This Paper Address?

Seismic data analysis is notoriously complex, and while wavelet transforms have offered significant improvements in data representation and restoration, traditional methods struggle with efficiently handling directional features and high-dimensional data, especially in noisy environments. Existing approaches often fail to adequately represent singularities like lines or planes, which are crucial for understanding subsurface structures. Furthermore, coherent noise, such as ground-roll, can obscure important geological information, and traditional filtering methods often fall short in complex media where this noise exhibits local directional changes.

Methods

The core contribution of this paper is the application of dual-tree M-band wavelet transforms to seismic data. This approach builds upon existing wavelet theory by incorporating two key advancements:

  • Orthogonal M-band filter banks: These are designed to better separate frequency bands in seismic data compared to standard wavelets. The increased degrees of freedom in filter design allow for more precise frequency partitioning.
  • Hilbert transform and complex representation: By using a Hilbert transform to create a complex representation of seismic signals, the method enhances the ability to define attributes and capture directional information more effectively.

The "dual-tree" aspect refers to the parallel processing of two wavelet trees, where the filters in one tree are related to the filters in the other by a Hilbert transform. This structure provides a degree of redundancy (specifically, a factor of two) that is significantly lower than some other overcomplete transforms, making it more computationally feasible for large datasets. The paper also briefly mentions the theoretical underpinnings of M-band multiresolution analysis and the design of dual-tree filters.

Datasets

The paper uses a seismic image for preliminary tests and a shot gather seismic dataset for the ground-roll filtering application. These appear to be real-world seismic data, though specific details about their origin or creation are not provided beyond their general description. No new benchmarks are introduced.

Key Insights and Ideas

  • Enhanced Directionality: The dual-tree M-band wavelet transform, particularly through its Hilbert pair structure, offers improved directional analysis capabilities. This allows for better identification and separation of features with specific orientations within the seismic data.
  • Effective Noise Removal: The method demonstrates success in removing coherent noise, such as ground-roll, even when this noise exhibits local directional variations. This is a significant advantage over traditional methods that rely on global velocity assumptions.
  • Balancing Redundancy and Performance: The dual-tree approach achieves a useful level of redundancy (factor of two) that aids in directional analysis and noise suppression without leading to the prohibitive computational costs associated with highly overcomplete transforms.
  • Revealing Hidden Information: By effectively filtering out noise, the technique can reveal subtle geological reflection signals that were previously obscured, leading to a clearer understanding of subsurface structures.

Key Results

The paper presents qualitative results demonstrating the effectiveness of the dual-tree M-band wavelet transform for:

  • Local Slope Analysis: Figure 4 shows that energy maps generated by the transform highlight areas where the seismic data's frequency content and slope match the characteristics of specific wavelets, indicating successful local directional analysis.
  • Ground-Roll Filtering: Figure 5 illustrates the removal of directional noise (ground-roll) from a seismic shot gather. The "filtered residual" image shows that important reflection signals emerge after the noise has been suppressed, suggesting that the method can uncover previously hidden geological information.

Quantitative results (e.g., signal-to-noise ratio improvement, specific accuracy metrics) are not explicitly detailed in the provided text.

Limitations and Future Work

  • Limitations: The paper acknowledges that while the dual-tree transform offers improvements, it still has limitations. The "infinite support of curvelets and their redundancy may limit their performance in some applications" is mentioned as a point of comparison, implying that while the dual-tree transform is better in this regard, it's not a perfect solution. The computational cost, though reduced compared to some methods, is still a factor.
  • Future Work: The authors suggest that future work will focus on the application of this method for seismic data compression.

Summary updated

Abstract

Seismic Data Analysis With Dual-Tree M-Band Wavelet Transforms Laurent Duval (Institut Français du Pétrole (IFP)) Caroline Chaux (ARIANA research group) and Stéphan Ker (IFREMER Marine Geosciences Department) SUMMARY____________________________________________________________ The complexity of seismic data still challenges signal processing algorithms in several applications. The rediscovery of wavelet transforms by J.Morlet et al. has allowed improvements in addressing several data representation (local analysis compression) and restoration problems. However despites their achievements traditional approaches based on discrete and separable (both for computational purposes) wavelets fail at efficiently representing directional or higher dimensional data features such as line or plane singularities especially in severe noise
BibTeX
@misc{duval2007seismic,
  author = {Laurent Duval and Caroline Chaux and Stéphan Ker},
  title = {{Seismic data analysis with dual-tree M-band wavelet transforms}},
  year = {2007},
  doi = {10.3997/2214-4609.201405088},
}
Figure